CN113382087A

CN113382087A - Push time prediction method, device, equipment, storage medium and program product

Info

Publication number: CN113382087A
Application number: CN202110791819.4A
Authority: CN
Inventors: 黄福华; 郑文琛
Original assignee: WeBank Co Ltd
Current assignee: WeBank Co Ltd
Priority date: 2021-07-13
Filing date: 2021-07-13
Publication date: 2021-09-10

Abstract

The invention discloses a push time prediction method, a push time prediction device, a push time prediction equipment, a storage medium and a program product, wherein the method comprises the following steps: predicting by adopting a frequency model based on the target user characteristics of the target user and preset alternative push frequencies to obtain a first prediction result corresponding to the alternative push frequencies; determining a target push frequency from the alternative push frequencies according to the first prediction result, and determining a push period according to the target push frequency; predicting by adopting a time model based on the characteristics of the target user, preset push content and each alternative time in the push period to obtain a second prediction result corresponding to each alternative time; and determining target push time for pushing preset push content to the target user in the push period from the alternative times according to the second prediction result. The invention realizes the pushing to the user with the pushing frequency and the pushing time which can be accepted by the user, and improves the experience degree of the user for acquiring the information.

Description

Push time prediction method, device, equipment, storage medium and program product

Technical Field

The present invention relates to the field of artificial intelligence technologies, and in particular, to a push time prediction method, apparatus, device, storage medium, and program product.

Background

At present, with the development of network technology, more and more channels and ways for users to acquire information contents are provided, and a way for pushing information contents such as articles and products to users based on pushing subjects such as public numbers and network shops is also gradually a popular channel for users to acquire information. However, when the push agent pushes content to the user, the push may be too frequent and the push time is not appropriate, which may cause the user to feel a sense of discomfort, and thus the user experience of obtaining information is poor.

Disclosure of Invention

The invention mainly aims to provide a push time prediction method, a push time prediction device, push time prediction equipment, a storage medium and a program product, and aims to solve the technical problem that when a current push main body pushes contents to a user, the user feels dislike due to too frequent pushing and improper push time, so that the experience of obtaining information by the user is poor.

In order to achieve the above object, the present invention provides a push time prediction method, including the following steps:

predicting by adopting a frequency model based on target user characteristics of a target user and preset alternative push frequencies to obtain a first prediction result corresponding to the alternative push frequencies, wherein the frequency model is obtained by training by taking the user characteristics and the push frequencies as input characteristics in advance and taking the attention state of the user to a push main body as a label;

determining a target push frequency from the alternative push frequencies according to the first prediction result, and determining a push period according to the target push frequency;

predicting by adopting a time model based on the target user characteristics, preset push contents and each alternative time in the push period to obtain a second prediction result corresponding to each alternative time, wherein the time model is obtained by taking the user characteristics, the push contents and the push time as input characteristics in advance and training by taking the attention state of a user to the push main body as a label;

and determining target push time for pushing the preset push content to the target user in the push cycle from each alternative time according to the second prediction result.

Further, the alternative pushing frequency is characterized as a vector, elements in the vector respectively correspond to each time period in a preset time period and are used for characterizing whether pushing is needed in the corresponding time period, the first prediction result represents a probability of canceling attention of the target user to the pushing subject when the target user is pushed according to the corresponding alternative pushing frequency,

the step of determining a target push frequency from the alternative push frequencies according to the first prediction result comprises:

and selecting a target pushing frequency from the alternative pushing frequencies according to the first prediction result, wherein the total pushing times of the target pushing frequency in the preset time period are the most in all qualified pushing frequencies, and the qualified pushing frequencies are the alternative pushing frequencies of which the corresponding attention cancellation probability is smaller than a preset threshold value in the alternative pushing frequencies.

Further, the step of determining a push period according to the target push frequency includes:

and taking a time period corresponding to the element which is represented to be pushed in the vector of the target pushing frequency as a pushing period.

Further, the step of predicting by using a time model based on the target user characteristic, the preset push content, and each alternative time in the push cycle to obtain a second prediction result corresponding to each alternative time includes:

arranging and combining the alternative pushed contents and the alternative times in the pushing period to obtain content time combinations;

inputting each content time combination and the target user characteristics into the time model for prediction to obtain a second prediction result corresponding to each content time combination, wherein the second prediction result represents the attention cancellation probability of the target user to the pushing main body when the target user pushes the alternative pushing content in the content time combination to the target user at the alternative time in the content time combination;

the step of determining, from the alternative times according to the second prediction result, a target push time for pushing the preset push content to the target user within the push cycle includes:

selecting a target content time combination with the minimum corresponding attention cancellation probability from the content time combinations according to the second prediction result;

and taking the alternative time in the target content time combination as the target pushing time for pushing the alternative pushing content in the target content time combination to the target user in the pushing period.

Further, before the step of predicting based on the target user characteristics of the target user and preset alternative push frequencies by using the frequency model to obtain the first prediction result corresponding to each alternative push frequency, the method further includes:

respectively determining sample users from the pushed users of the current pushing after each pushing, wherein the sample users are at least from the pushed users who cancel attention to the pushing main body within a first preset time length after the current pushing, and the pushed users who do not cancel attention to the pushing main body and have access records to the pushing main body within the first preset time length after the current pushing;

acquiring user characteristics of the sample user and pushing frequency of the sample user in a second preset time before the current pushing as input characteristics, and generating first sample data corresponding to the sample user by taking a concerned state of the sample user to the pushing main body in the first preset time after the current pushing as a label;

and training by using the first sample data corresponding to each sample user to obtain the frequency model.

Further, after the step of determining a sample user from the current pushed users after each content push, the method further includes:

acquiring the user characteristics of the sample user, the current pushing time and the current pushing content as input characteristics, and generating second sample data corresponding to the sample user by taking the attention state of the sample user to the pushing main body within the first preset time after the sample user is pushed for the current time as a label;

and training by adopting the second sample data corresponding to each sample user to obtain the time model.

Further, the step of collecting the user characteristics of the sample user comprises:

acquiring the times of accessing the pushing main body by the sample user within a third preset time before the current pushing as the user characteristics of the sample user; and/or the presence of a gas in the gas,

and acquiring a time interval between access time and current push time as the user characteristics of the sample user, wherein the access time is the access time of the sample user accessing the push subject within the third preset time before the current push, or the access time of the sample user accessing the push subject for the latest time before the current push.

To achieve the above object, the present invention further provides a push time prediction apparatus, including:

the first prediction module is used for predicting based on target user characteristics of a target user and preset alternative push frequencies by adopting a frequency model to obtain a first prediction result corresponding to each alternative push frequency, wherein the frequency model is obtained by training by taking the user characteristics and the push frequencies as input characteristics in advance and taking the attention state of the user to a push main body as a label;

a first determining module, configured to determine a target push frequency from the alternative push frequencies according to the first prediction result, and determine a push period according to the target push frequency;

the second prediction module is used for predicting based on the target user characteristics, preset push contents and each alternative time in the push period by adopting a time model to obtain a second prediction result corresponding to each alternative time, wherein the time model is obtained by training by taking the user characteristics, the push contents and the push time as input characteristics in advance and taking the attention state of the user to the push main body as a label;

a second determining module, configured to determine, according to the second prediction result, a target push time for pushing the preset push content to the target user within the push cycle from each of the candidate times.

To achieve the above object, the present invention also provides a push time prediction apparatus, including: a memory, a processor and a push time prediction program stored on the memory and executable on the processor, the push time prediction program when executed by the processor implementing the steps of the push time prediction method as described above.

Furthermore, to achieve the above object, the present invention also proposes a computer readable storage medium having stored thereon a push time prediction program, which when executed by a processor, implements the steps of the push time prediction method as described above.

Furthermore, to achieve the above object, the present invention also proposes a computer program product comprising a computer program which, when being executed by a processor, implements the steps of the push time prediction method as described above.

In the invention, a frequency model is adopted to predict based on the target user characteristics of a target user and each alternative push frequency to obtain a first prediction result corresponding to each alternative push frequency, because the frequency model is obtained by training by taking the user characteristics and the push frequency as input characteristics and the attention state of the user to the push main body as a label in advance, the user's acceptance of the push frequency can be predicted, so that a target push frequency with a high acceptance by the target user can be determined from the various alternative push frequencies according to the first prediction result, thereby making it easier for the target user to accept when pushing content to the target user according to the push period determined according to the target push frequency, therefore, the information acquisition experience of the user is improved, the attention cancellation rate of the pushing main body can be reduced, and the continuous attention of the user to the pushing main body is kept. And a time model is adopted to predict based on the target user characteristics, the preset push content and each alternative time in the push period to obtain a second prediction result corresponding to each alternative time, because the time model is obtained by training by taking the user characteristics, the push content and the push time as input characteristics and taking the attention state of the user to the push subject as a label in advance, the user's acceptance of pushing the content at that time can be predicted, and therefore a target push time that is accepted to a high degree by the target user can be determined from among the respective candidate times based on the second prediction result, thereby, when the preset push content is pushed to the target user at the target push time in the push period, the target user can accept the preset push content more easily, therefore, the information acquisition experience of the user is further improved, and the attention cancellation rate of the pushing main body is further reduced.

Drawings

FIG. 1 is a schematic diagram of a hardware operating environment according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a first embodiment of a push time prediction method according to the present invention;

fig. 3 is a schematic view of a usage scenario of a frequency model according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a usage scenario of a time model according to an embodiment of the present invention;

fig. 5 is a functional block diagram of a push time prediction apparatus according to a preferred embodiment of the present invention.

The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As shown in fig. 1, fig. 1 is a schematic device structure diagram of a hardware operating environment according to an embodiment of the present invention.

It should be noted that, the push time prediction device in the embodiment of the present invention may be a smart phone, a personal computer, a server, and the like, and is not limited herein.

As shown in fig. 1, the push time prediction apparatus may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a storage device separate from the processor 1001.

Those skilled in the art will appreciate that the device configuration shown in fig. 1 does not constitute a limitation of the push time prediction device and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.

As shown in fig. 1, a memory 1005, which is a kind of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and a push time prediction program. The operating system is a program that manages and controls the hardware and software resources of the device, supporting the operation of the push time prediction program as well as other software or programs. In the device shown in fig. 1, the user interface 1003 is mainly used for data communication with a client; the network interface 1004 is mainly used for establishing communication connection with a server; and the processor 1001 may be configured to invoke the push time prediction program stored in the memory 1005 and perform the following operations:

the determining a target push frequency from the alternative push frequencies according to the first prediction result comprises:

Further, the determining a push period according to the target push frequency includes:

Further, the preset push content includes a plurality of candidate push contents, and the predicting, by using a time model, based on the target user characteristic, the preset push content, and each candidate time in the push cycle, to obtain a second prediction result corresponding to each candidate time includes:

the determining, from the alternative times according to the second prediction result, a target push time for pushing the preset push content to the target user within the push cycle includes:

Further, before the adopted frequency model predicts based on the target user characteristics of the target user and preset alternative push frequencies to obtain the first prediction result corresponding to the alternative push frequencies, the processor 1001 may be further configured to call a push time prediction program stored in the memory 1005, and perform the following operations:

Further, after determining the sample user from the current pushed users after pushing the content each time, the processor 1001 may be further configured to invoke a push time prediction program stored in the memory 1005, and perform the following operations:

Further, the collecting user characteristics of the sample user comprises:

Based on the above structure, various embodiments of the push time prediction method are proposed.

Referring to fig. 2, fig. 2 is a flowchart illustrating a push time prediction method according to a first embodiment of the present invention.

While a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in an order different than that shown. In this embodiment, the push time prediction method execution subject may be a device such as a smart phone, a personal computer, a server, etc., and for convenience of description, the push system is taken as the execution subject to describe the embodiments. In this embodiment, the push time prediction method includes:

step S10, predicting by using a frequency model based on target user characteristics of a target user and preset alternative push frequencies to obtain a first prediction result corresponding to the alternative push frequencies, wherein the frequency model is obtained by training by taking the user characteristics and the push frequencies as input characteristics in advance and taking the attention state of the user to a push subject as a label;

the frequency model can be preset in the push system in advance, and can be used for predicting the acceptance degree of the user to various push frequencies, namely, the frequency model can be used for predicting to obtain a prediction result capable of reflecting the acceptance degree of the user to the push frequencies based on the user characteristics of the user and the push frequencies. The frequency model can be obtained by pre-training, the user characteristic and the push frequency are used as input characteristics of the frequency model, and the attention state of the user to the push subject is used as a training label. The user characteristics may include user portrait characteristics such as age, gender, and/or number of accesses to the push subject of the user, which is not limited in this embodiment. The push frequency represents the frequency of pushing the content to the user within a period of time, and in this embodiment, the data format of the push frequency is not limited as long as the push frequency can be reflected; for example, the pushing frequency may specifically be a numerical value indicating the number of times of pushing in the period of time, or may be a plurality of numerical values respectively indicating whether to push in each time period of the period of time. The frequency model may adopt a two-class or multi-class model structure, such as a linear model like LR (logistic regression), a tree model like GBDT (gradient lifting tree), or a neural network model like DNN (deep neural network), and specifically which model structure is adopted is not limited in this embodiment; the training method of the frequency model can adopt a conventional supervised training method of a machine learning model, and detailed description is not repeated herein; the training process of the frequency model can be executed by the push system, and the frequency model can be obtained by training of other equipment and then deployed into the push system. The push agent is an agent that pushes content to a user, such as a public number, a network store, and the like, and the push content may be various contents such as articles, commodities, music, and the like, and is not limited in this embodiment. The operation and maintenance personnel of the pushing main body can register the pushing main body in the pushing system and upload the pushing content, and the pushing system executes a pushing process of pushing the content to the user on the behalf of the pushing main body.

The attention state represents a user response when the user pushes content according to a certain push frequency, for example, after pushing according to the push frequency, the user cancels attention, which indicates that the user is likely not to accept the push frequency, and if continuing to pay attention, indicates that the user can accept the push frequency, so that the user characteristic and the push frequency are used as input characteristics, and the attention state of the user to the push main body is used as a frequency model obtained by label training, so that the acceptance degree of the user to the push frequency can be predicted, or the possibility that the user cancels attention to the push main body after pushing content to the user according to the push frequency can be predicted.

The data format of the prediction result obtained by the frequency model prediction is not limited in this embodiment. In one embodiment, a result indicating whether the user accepts (or cancels the attention) may be directly output, for example, output 1 indicates acceptance and output 0 indicates no acceptance. In another embodiment, it may be arranged to output a probability value; according to different requirements of specific scenes, a probability value output by a frequency model after prediction is performed based on a user characteristic of a certain user and a certain pushing frequency can be set to represent the probability that the user cancels attention to a pushing subject or the probability that the user does not cancel attention after the user is pushed according to the pushing frequency, and in any case, the probability value can obviously reflect the acceptance degree of the user to the pushing frequency. In other embodiments, other forms of prediction results may also be output. It should be noted that the output probability value represents the probability of canceling attention or the probability of not canceling attention can be realized by different setting modes of the tag; for example, when the output probability value represents a probability of canceling attention, a label of canceling attention may be set to "1" and a label of not canceling attention may be set to "0", whereas when the output probability value represents a probability of not canceling attention, a label of canceling attention may be set to "0" and a label of not canceling attention may be set to "1".

When the push time for pushing the content to the target user needs to be determined, the push system may perform prediction based on the user characteristics of the target user (hereinafter, referred to as target user characteristics for distinction) and each alternative push frequency by using a frequency model, so as to obtain a prediction result (hereinafter, referred to as a first prediction result for distinction) corresponding to each alternative push frequency. Wherein, a plurality of different alternative push frequencies can be preset to represent different frequency degrees of push contents. For example, when the push frequency refers to the number of times of pushing within one month, a plurality of frequencies such as 1 time, 2 times, 3 times, and the like may be set as the alternative recommendation frequency. When the frequency model adopts a binary model structure, the push system can respectively predict the target user characteristics and each alternative push frequency input frequency model, that is, the target user characteristics and one alternative push frequency are input each time, and a first prediction result corresponding to the alternative push frequency is obtained. When the frequency model adopts a multi-classification model structure, the pushing system can input the target user characteristics and all the alternative pushing frequencies into the frequency model together for prediction to obtain a first prediction result corresponding to each alternative pushing frequency.

Step S20, determining a target push frequency from the alternative push frequencies according to the first prediction result, and determining a push period according to the target push frequency;

and the push system determines a target push frequency from the alternative push frequencies according to the first prediction result. Specifically, since the first prediction results may reflect the acceptance degree of the target user for the alternative push frequency, the target push frequency may be selected from the alternative push frequencies according to the acceptance degree reflected by each first prediction result. For example, in an embodiment, the push system may use a push frequency with the highest acceptance degree of the target user in the alternative push frequencies as the target push frequency, and in other embodiments, the target push frequency may also be selected in combination with the size of the push frequency, which is not limited in this embodiment.

After determining the target push frequency, the push system may determine a push period according to the target push frequency. Specifically, a push period refers to a time span of pushing content once, that is, the content is pushed only once in one push period. In a specific embodiment, there may be one or more push periods; if there is one, the push period is a period of time counted from the push start time, e.g., if the target push frequency is 4 times every 28 days, the first push period may be 7 days counted from the push start time; the push starting time is the current time or a preset starting time, for example, if a push plan is scheduled to be implemented starting from 1 month and 1 day, the push starting time is 1 month and 1 day; if there are multiple push cycles, each push cycle is multiple consecutive time periods counted from the push start time, for example, if there are four push cycles and the target push frequency is 4 times every 28 days, the first push cycle is 7 days counted from the push start time, the second push cycle is seven days counted from the first push cycle end time, the third push cycle is seven days counted from the second push cycle end time, and so on.

Step S30, predicting by adopting a time model based on the target user characteristics, preset push contents and each alternative time in the push period to obtain a second prediction result corresponding to each alternative time, wherein the time model is obtained by training by taking the user characteristics, the push contents and the push time as input characteristics in advance and taking the attention state of the user to the push subject as a label;

the recommendation system may preset a time model, and the time model may be used to predict the degree of acceptance of the user on the push content and the push time, that is, a prediction result capable of reflecting the degree of acceptance of the user on the push content and the push time may be predicted based on a user characteristic of the user, the push content, and the push time by using the time model. The time model can be obtained by training in advance, the user characteristics, the push content and the push time are used as input characteristics of the time model, and the attention state of the user to the push subject is used as a training label. The user characteristics may include user portrait characteristics such as age and gender of the user and/or the number of accesses to the push subject, and may have the same or different characteristic dimensions as the user characteristics used in the frequency model, which is not limited in this embodiment. The push content can be contents in various forms such as texts, images, audio and the like, and can be converted into a form of a feature vector as an input feature in a feature mapping mode; for example, text is converted into vectors by a Word2vec method, images are converted into feature vectors by an image feature extraction model, and the like. The pushing time may be represented in the form of time with different time granularities, such as year, month, and day, or may be represented in the form of day in month and day in week, which is not limited in this embodiment. The time model may adopt a two-classification or multi-classification model structure, such as a linear model like LR, a tree model like GBDT, or a neural network model like DNN, and specifically which model structure is adopted is not limited in this embodiment; the training method of the time model can adopt a conventional supervised training method of a machine learning model, and detailed description is not repeated herein; the training process of the time model can be executed by the push system, and the time model can be obtained by training of other equipment and then deployed into the push system.

The attention state represents a user's reaction when the user pushes a certain content at a certain push time according to a certain push frequency to the push subject, for example, after the content is pushed to the user at the push time according to the push frequency, the user cancels attention, which indicates that the user is likely not to accept pushing the content at the push time, if the user continues to pay attention, the user can accept pushing the content at the push time, so the user characteristic, the push content and the push time are used as input characteristics, and the attention state of the user to the push subject is used as a time model obtained by tag training, which can predict the acceptance degree of the user to push the content at the time, or predict the possibility that the user cancels attention to the push subject after pushing the content to the user at the push time.

The present embodiment does not limit the data format of the prediction result obtained by the time model prediction. In one embodiment, a result indicating whether the user accepts (or cancels the attention) may be directly output, for example, output 1 indicates acceptance and output 0 indicates no acceptance. In another embodiment, it may be arranged to output a probability value; according to different requirements of specific scenes, the probability value output by the time model after prediction is performed based on the user characteristics of a certain user, a certain pushing time and a certain pushing content can be set to represent the probability that the user cancels attention to the pushing subject or the probability that the user does not cancel attention after the pushing content is pushed to the user at the pushing time, and in any case, the probability value can obviously reflect the acceptance degree of the user to the pushing content and the pushing time. In other embodiments, other forms of prediction results may also be output. It should be noted that the output probability value represents the probability of canceling attention or the probability of not canceling attention can be realized by different setting modes of the tag; for example, when the output probability value represents a probability of canceling attention, a label of canceling attention may be set to "1" and a label of not canceling attention may be set to "0", whereas when the output probability value represents a probability of not canceling attention, a label of canceling attention may be set to "0" and a label of not canceling attention may be set to "1".

After the push period is determined, for each push period, the push system may predict, by using a time model, based on the target user characteristics, the preset push content, and each alternative time in the push period, to obtain a prediction result (hereinafter referred to as a second prediction result for illustration) corresponding to each alternative time. It should be noted that when there are a plurality of preset push contents, there are a plurality of second prediction results corresponding to one candidate time, that is, the second prediction results corresponding to the candidate time when the candidate time is respectively combined with each push content are included. For example, if a push cycle is seven days from the push start time, the candidate times in the push cycle may be seven, which are monday, tuesday, wednesday, and thursday … … sunday of the seven days. When the time model adopts a binary model structure, the push system can respectively predict the target user characteristics and the preset push contents with each alternative time input time model, that is, the target user characteristics, the preset push contents and one alternative time are input each time, and a second prediction result corresponding to the alternative time is obtained. When the time model adopts a multi-classification model structure, the pushing system can input the target user characteristics, the preset pushing content and each alternative time into the time model together for prediction to obtain a second prediction result corresponding to each alternative time.

Step S40, determining, according to the second prediction result, a target push time for pushing the preset push content to the target user in the push cycle from each of the candidate times.

The push system may determine, according to the second prediction result, a target push time for pushing the preset push content to the target user within the push period from among the candidate times. Specifically, since the second prediction results may reflect the acceptance degree of the target user for the alternative time and the push content, the target push time may be selected from the alternative time according to the acceptance degree reflected by each second prediction result. For example, in an embodiment, when there is one preset push content, the push system may use a time with the highest acceptance of the target user in the alternative times as a target push time for pushing the preset push content to the user. For another example, in another embodiment, when there are a plurality of preset push contents, a second prediction result corresponding to each push content at each alternative time is obtained through time prediction, and for each push content, a time with the highest target user acceptance degree in the alternative times may be used as a target push time for pushing the preset push content to the user; or, when there are a plurality of preset push contents but only one push content is finally needed to be pushed to the user, a combination of the alternative time with the highest user acceptance degree and the push content may be selected, and the alternative time is used as the push time for pushing the push content.

After the target push time is determined, the push system can push preset push content to the target user according to the target push time.

In the embodiment, a frequency model is adopted to predict based on the target user characteristics of the target user and each alternative push frequency to obtain a first prediction result corresponding to each alternative push frequency, because the frequency model is obtained by training by taking the user characteristics and the push frequency as input characteristics and the attention state of the user to the push main body as a label in advance, the user's acceptance of the push frequency can be predicted, so that a target push frequency with a high acceptance by the target user can be determined from the various alternative push frequencies according to the first prediction result, thereby making it easier for the target user to accept when pushing content to the target user according to the push period determined according to the target push frequency, therefore, the information acquisition experience of the user is improved, the attention cancellation rate of the pushing main body can be reduced, and the continuous attention of the user to the pushing main body is kept. And a time model is adopted to predict based on the target user characteristics, the preset push content and each alternative time in the push period to obtain a second prediction result corresponding to each alternative time, because the time model is obtained by training by taking the user characteristics, the push content and the push time as input characteristics and taking the attention state of the user to the push subject as a label in advance, the user's acceptance of pushing the content at that time can be predicted, and therefore a target push time that is accepted to a high degree by the target user can be determined from among the respective candidate times based on the second prediction result, thereby, when the preset push content is pushed to the target user at the target push time in the push period, the target user can accept the preset push content more easily, therefore, the information acquisition experience of the user is further improved, and the attention cancellation rate of the pushing main body is further reduced.

Further, based on the first embodiment, a second embodiment of the push time prediction method according to the present invention is proposed, and in this embodiment, the step of determining the target push frequency from the alternative push frequencies according to the first prediction result in step S20 includes:

step S201, selecting a target push frequency from each of the alternative push frequencies according to the first prediction result, where a total number of times of pushing of the target push frequency in the preset time period is the largest among all qualified push frequencies, and the qualified push frequency is an alternative push frequency in each of the alternative push frequencies, where a corresponding attention cancellation probability is smaller than a preset threshold.

In this embodiment, the alternative push frequency may be characterized as a vector, that is, the data format of the alternative push frequency is a vector, and elements in the vector respectively correspond to time periods within a preset time period and are used for characterizing whether push is required within the corresponding time period. The preset time period may be a time period with a certain duration from the push start time, for example, if the push start time is 1 month and 1 day, and the duration of the preset time period is set to be 30 days, the preset time period is 1 month and 1 day to 1 month and 30 days. The number of the time periods in the preset time period may be set in advance as needed, for example, four time periods are set, the preset time period is divided into four time periods, for example, the preset time period is 1 month and 1 day to 1 month and 30 days, and the four time periods are four weeks from 1 month and 1 day to 1 month and 30 days, respectively (if the time period relates to five weeks, incomplete weeks may be merged into other weeks). The elements in the vector correspond to respective time periods within a preset time period, specifically, the elements in the vector correspond to the time periods one by one, for example, if there are 4 time periods, there are four elements in the vector. The value of the element is used to indicate whether pushing is needed in the corresponding time period, for example, a value of 1 of the element indicates that pushing is needed in the corresponding time period, and a value of 0 indicates that pushing is not needed in the corresponding time period, and different combinations of values of the elements in the vector, that is, different alternative pushing frequencies, are used. It can be understood that when pushing is required in each time period, the total number of times of pushing in the preset time period is the largest, which means that pushing is the most frequent, and when only one time period requires pushing, the total number of times of pushing in the preset time period is the smallest, which means that pushing is the least frequent; the total number of times of pushing of the two alternative pushing frequencies in the preset time period may be the same, for example, when the value of the element is 1, which indicates that pushing is required in the corresponding time period, and 0, which indicates that pushing is not required in the corresponding time period, if there are four vector elements and the vectors of the two alternative pushing frequencies are [1,1,1,0] and [1,1,0,1], the total number of times of pushing of the two alternative pushing frequencies is 3, the difference is that the first alternative pushing frequency needs to be pushed in the third period and not pushed in the fourth period, and the second alternative pushing frequency needs to be pushed in the third period and fourth period.

It should be noted that, corresponding to the data format of the alternative push frequency, the push frequency input during the training of the frequency model is also the same data format, and specifically, a vector may be used to indicate whether to push to the user in each period of the past period.

In this embodiment, the data format of the prediction result obtained by the frequency model prediction may be set as a probability value, and the probability value may represent the probability that the user pays no attention to the push subject. Specifically, the probability value may be set as a probability that the user cancels the attention of the push subject, or may be set as a probability that the user does not cancel the attention of the push subject, but either of the probabilities may represent the probability that the user cancels the attention of the push subject; that is, when the probability value is set to the probability that the user cancels the attention to the push subject, the size of the probability value directly reflects the probability that the user cancels the attention to the push subject, and when the probability value is set to the probability that the user does not cancel the attention to the push subject, the size obtained by the 1-probability value reflects the probability that the user cancels the attention to the push subject. Then, the first prediction result represents a concern canceling probability of the target user to the pushing subject when pushing is performed on the target user according to the corresponding alternative pushing frequency.

The push system can select a target push frequency from the alternative push frequencies according to a first prediction result after obtaining the first prediction result corresponding to each alternative push frequency, wherein the total push times of the selected target push frequency in a preset time period are the largest in all qualified push frequencies, and the qualified push frequencies refer to the alternative push frequencies of which the corresponding attention cancellation probabilities are smaller than a preset threshold value. That is, if the attention cancellation probability corresponding to one candidate push frequency is smaller than the preset threshold, the candidate push frequency belongs to the qualified push frequencies, and if the total push times of one qualified push frequency in the preset time period are the largest among the qualified push frequencies, the qualified push frequency is the target push frequency. The preset threshold may be a threshold that is set in advance according to needs, for example, set to 0.5. The total pushing times of the alternative pushing frequency in the preset time period are the number of time periods needing to be pushed, which is determined according to each element of the pushing frequency.

Specifically, in an embodiment, the push system may calculate a cancellation attention rate corresponding to each alternative push frequency according to the first prediction result, compare the cancellation attention rate of each push frequency with a preset threshold value, select a qualified push frequency, compare the total push times of each qualified push frequency, and select the qualified push frequency with the largest total push time as the target push frequency.

In another embodiment, the push system may traverse the alternative push frequencies in order from high to low according to the total number of pushes, and take the traversed first qualified push frequency as the target push frequency. For example, as shown in fig. 3, the alternative push frequencies are listed under the push frequency (the tweet frequency feature in the corresponding graph), and are arranged in sequence from high to low according to the total number of times of push, and the alternative push frequencies and the user feature are respectively combined and input into the frequency model for prediction, so as to obtain first prediction results (prediction scores in the corresponding graph, and the first prediction results are not separately represented in the graph) corresponding to the alternative push frequencies, and the alternative push frequencies are traversed from top to bottom, and the first qualified push frequency is taken as the target push frequency. For example, when the preset threshold is 0.5, if the attention cancellation probability corresponding to the alternative push frequency [1,1,1,1] is 0.6, and the attention cancellation probability corresponding to [1,1,1,0] is 0.4, then with [1,1,1,0] as the target push frequency, the push cycle determined according to the target push frequency is 1, 2, 3 weeks, and the push is not performed in 4 weeks.

Further, in an embodiment, the step of determining a push cycle according to the target push frequency in step S20 includes:

step S202, a time period corresponding to an element representing that the element needs to be pushed in the vector of the target pushing frequency is used as a pushing period.

After determining the target push frequency, the push system can search for an element which is represented in the vector of the target push frequency and needs to be pushed in a corresponding time period, and the time period corresponding to the element is used as a push period. For example, when the value of an element is 1 indicating that push is needed, and 0 indicating that push is not needed, if there are four periods in total, and there are four elements in the vector of the target push frequency, which are 1, and 0, respectively, the push system determines that the first period, the second period, and the third period are push periods, and the fourth period is not push.

In this embodiment, a vector representing whether to perform pushing in each time period is used to represent the pushing frequency, and compared with a simple method of representing the pushing frequency by using the number of times, the frequency of pushing in a period of time can be represented more finely, for example, [1,1,1,0] and [1,1,0,1] where the pushing number of times is three correspond to different pushing frequencies in this embodiment, so that the receiving degrees of the user to different pushing frequencies can be located more accurately, content can be pushed to the user by using the more accurate pushing frequency, the experience of the user in acquiring information is further improved, and the attention cancellation rate of the pushing main body is further reduced.

Further, based on the first and/or second embodiments, a third embodiment of the push time prediction method of the present invention is proposed, in this embodiment, the step S30 includes:

step S301, arranging and combining each alternative push content and each alternative time in the push period to obtain each content time combination;

in this embodiment, the preset push content may include a plurality of candidate push contents, that is, a plurality of candidate push contents may be preset, and the most suitable push content is selected to be pushed to the user at the most suitable time according to a prediction result of the time model. When there are multiple push periods, the alternative push contents corresponding to each push period may be the same or different.

After the push system determines the push period, taking a push period as an example, the push system may arrange and combine each alternative push content and each alternative time in the push period to obtain each content time combination. For example, as shown in fig. 4, the alternative times (corresponding to the tweet time feature in the figure) in the push cycle include week 1, week 2, and week 3 … … sunday (not shown in the week 3-sunday figure), the alternative push contents (corresponding to the tweet content feature in the figure) include text 1 and text 2, and each alternative time is combined with text 1 and text 2, respectively, so as to obtain each content time combination, and fig. 4 shows four content time combinations "week 1-text 1", "week 1-text 2", "week 2-text 1", and "week 2-text 2".

Step S302, inputting each content time combination and the target user characteristics into the time model for prediction to obtain a second prediction result corresponding to each content time combination, wherein the second prediction result represents the attention cancellation probability of the target user to the pushing subject when the target user pushes the alternative pushing content in the content time combination to the target user at the alternative time in the content time combination;

after each content time combination is obtained, the pushing system outputs the content time combination and the target user characteristics together to a time model for prediction, and a second prediction result corresponding to each content time combination is obtained. As shown in fig. 4, one line represents that the data of the time model is input once, and the second prediction results corresponding to the respective content time combinations are output (corresponding to the prediction scores in the figure, the first prediction results are not separately shown in the figure).

The step S40 includes:

step S401, selecting a target content time combination with the minimum corresponding attention cancellation probability from the content time combinations according to the second prediction result;

in this embodiment, the data format of the prediction result obtained by the time model prediction may be set as a probability value, and the probability value may represent the probability that the user pays no attention to the push subject. Specifically, the probability value may be set as a probability that the user cancels the attention of the push subject, or may be set as a probability that the user does not cancel the attention of the push subject, but either of the probabilities may represent the probability that the user cancels the attention of the push subject; that is, when the probability value is set to the probability that the user cancels the attention to the push subject, the size of the probability value directly reflects the probability that the user cancels the attention to the push subject, and when the probability value is set to the probability that the user does not cancel the attention to the push subject, the size obtained by the 1-probability value reflects the probability that the user cancels the attention to the push subject. Then, the second prediction result represents a probability of canceling attention of the target user to the push subject when the alternative push content in the content time combination is pushed to the target user at the alternative time in the content time combination.

The push system may select, as the target content time combination, the content time combination with the smallest corresponding attention cancellation probability from the content time combinations according to the second prediction result after obtaining the second prediction result corresponding to each content time combination.

Step S402, using the alternative time in the target content time combination as a target push time for pushing the alternative push content in the target content time combination to the target user in the push cycle.

The push system may use the alternative push content in the target content time combination as the target push content that needs to be pushed to the target user in the push period, and use the alternative time in the target content time combination as the target push time that needs to push the target push content to the user in the push period. After determining the target push time and the target push content, the push system may push the target push content to the target user at the target push time within the push period. For example, as in fig. 4, if the predicted cancellation attention rate is lowest for tweet time week 1 and tweet 2 plus the user feature input model, tweet 2 is scheduled to be pushed to the user on week 1.

In this embodiment, by setting a plurality of candidate push contents, combining the candidate push contents with each candidate time in a push cycle, predicting each content time combination by using a time model to obtain a second prediction result corresponding to each content time combination, taking the corresponding content time combination with the minimum attention cancellation probability as a target content time combination according to the second prediction result, and taking the candidate time in the target content time combination as the target push time for pushing the candidate push contents in the target content time combination to a target user in the push cycle, a push user can more accept the push contents at a push frequency more acceptable to the user at a push time more acceptable to the user is realized, the information acquisition experience of the user is further improved, and the attention cancellation rate of a push subject is further reduced.

Further, based on the first, second and/or third embodiments, a fourth embodiment of the push time prediction method of the present invention is provided, and in this embodiment, the method further includes:

step S50, respectively determining sample users from the pushed users of the current pushing after each pushing, wherein the sample users are at least from the pushed users who cancel attention to the pushing subject within a first preset time length after the current pushing, and the pushed users who do not cancel attention to the pushing subject and have access records to the pushing subject within the first preset time length after the current pushing;

in this embodiment, the push system may collect sample data used for training the frequency model in the previous push, and train to obtain the frequency model by using the sample data, and when performing subsequent push, may predict an appropriate push frequency based on the frequency model obtained by training, and perform push according to the push frequency.

In particular, the push system may determine a sample user from the pushed users of the current push after each push of the content. The push system pushes the content to a plurality of users each time, namely, the pushed users in the second push. The sample users are at least from pushed users who cancel attention to the pushing subject within a first preset time length after the pushing (hereinafter referred to as first type sample users) and pushed users who do not cancel attention to the pushing subject and have access records to the pushing subject within the first preset time length after the pushing (hereinafter referred to as second type sample users). That is, for each pushed user in one pushing, if the pushing system detects that the pushed user cancels the attention of the pushing subject within the first preset time after the pushing, the pushing system may use the pushed user as the sample user. For each pushed user in one pushing process, if the pushing system detects that the pushed user does not cancel attention to the pushing main body within the first preset time after the pushing process, and has an access record to the pushing main body within the first preset time, the pushing system may use the pushed user as a sample user. The first preset time period may be a time period preset according to needs, and may be set to one day, for example. In one embodiment, the first type sample user may be a positive sample user, and the second type sample user may be a negative sample user; in another embodiment, the second type sample user may be used as a positive sample user and the first type sample user may be used as a negative sample user.

In an embodiment, the number of the two types of sample users determined by the push system may be the same among the sample users determined after one push. Specifically, the number of the first type sample users may be used as a reference, and when the number of the second type sample users is less than the number of the first type sample users, the pushing system may also add, as the sample user, the pushed user who has not paid attention but has not accessed the record within a first preset time after pushing the text for the second time, to the second type sample user; when the number of the second type sample users is larger than that of the first type sample users, the push system may randomly sample the second type sample users, and extract the sample users with the same number as that of the first type sample users as final second type sample users. By ensuring that the number of the two types of sample users is the same, the proportion of positive and negative samples is close to 1:1, and therefore a model with high prediction accuracy is more favorably trained.

Step S60, collecting user characteristics of the sample user and the pushing frequency of the sample user in a second preset time before the current pushing as input characteristics, and generating first sample data corresponding to the sample user by using the attention state of the sample user to the pushing subject in the first preset time after the current pushing as a label;

after determining the sample user, the pushing system may generate a piece of sample data (hereinafter referred to as first sample data) for training the frequency model, where the sample data corresponds to the sample user, and the first sample data includes the input feature and the tag. The pushing system may collect the user characteristics of the sample user and the pushing frequency of the sample user within a second preset time period before the current pushing as the input characteristics in the first sample data. The user characteristics can be user portrait characteristics such as the age, sex and concerned pushing subject of the user, and the pushing system can acquire various characteristics of the user under the condition of user authorization. The push frequency of the sample user refers to the frequency of the push system for pushing the content to the sample user; the pushing frequency of the sample user in the second preset duration before the second pushing refers to a frequency of pushing the content to the sample user by the pushing system in the second preset duration in the past. The second preset time period can be set as required, for example, to one month; when the pushing frequency is expressed in a vector form, the length of the second preset time period is the same as the time period of the preset time period corresponding to the alternative pushing frequency, the pushing frequency of the collected sample user is also expressed in a vector form, the number of elements of the vector is the same as the number of elements of the vector of the alternative pushing frequency, and each element represents whether the sample user is pushed or not in each period of the past period. The pushing system may use the attention state of the sample user to the pushing subject within the first preset time after the current pushing as a tag in the first sample data, and it can be understood that the final attention state of the sample user within the first preset time after the current pushing is used as the tag, that is, the tag of the first type of sample user is to cancel attention, and the tag of the second type of sample user is not to cancel attention.

It is understood that if a user does not pay attention after a push, the push system determines the user as a sample user, and if the user is also the pushed user at the next push, the user may be regarded as the sample user again by the push system; since the push frequency and the label may be different in the first sample data generated twice according to the sample user, two pieces of first sample data corresponding to the sample user may exist in the push system.

Step S70, training the first sample data corresponding to each sample user to obtain the frequency model.

After the first sample data of each sample user is collected in each pushing, the pushing system can train by adopting each first sample data to obtain a frequency model. Specifically, it may be that the training is started when the data amount of the collected first sample data is larger than a certain amount, for example, larger than one thousand pieces. The frequency model training method includes selecting a model structure in advance, setting an input layer and an output layer in the model structure according to input features and labels, initializing each model parameter in the model to obtain a model to be trained, training the model to be trained by adopting first sample data, iteratively updating each model parameter in the model with the aim of improving prediction accuracy, ending training until a certain training stop condition is met, and taking the model to be trained with the finally determined model parameters as the frequency model. The training stopping condition may be set as required, for example, AUC (Area Under Curve, defined as the Area enclosed by coordinate axes Under the ROC Curve) may be used as an effect measurement index, the first sample data is divided into a training set and a test set, training is performed on the training set, testing is performed on the test set, and AUC is calculated, when the AUC is greater than a certain threshold, the offline evaluation is considered to pass, the training is ended, otherwise, the training is continued until the AUC satisfies the condition.

Further, in an embodiment, after the step S50, the method further includes:

step S80, collecting the user characteristics of the sample user, the push time of the current push and the push content of the current push as input characteristics, and generating second sample data corresponding to the sample user by taking the attention state of the sample user to the push subject within the first preset time after the current push as a label;

in this embodiment, the pushing system may collect sample data used for training the time model in pushing in a previous time, train the sample data to obtain the time model, predict appropriate pushing time based on the trained time model in subsequent pushing, and push according to the pushing time.

Specifically, after determining the sample user, the pushing system may generate a piece of sample data (hereinafter referred to as second sample data) for training the time model, where the sample data corresponds to the sample user, and the second sample data includes the input feature and the tag. The pushing system can acquire the user characteristics of the sample user, the pushing time of the current pushing and the pushing content of the current pushing as the input characteristics in the second sample data. The user characteristics in the first sample data and the second sample data corresponding to the same sample user may be the same. The pushing system can use the attention state of the sample user to the pushing subject within a first preset time after the sample user is pushed for the time as a label in second sample data; it can be understood that the labels in the first sample data and the second sample data corresponding to the same sample user are the same.

And step S90, training by using the second sample data corresponding to each sample user to obtain the time model.

After the second sample data of each sample user is collected in each pushing, the pushing system can adopt each second sample data to train to obtain a time model. Specifically, the training may be started when the data amount of the collected second sample data is larger than a certain amount, for example, larger than one thousand pieces. The process of training the time model by using the second sample data may refer to the training process of the frequency model, which is not described in detail herein.

Further, in an embodiment, the step of collecting the user characteristics of the sample user in step S60 includes:

step S601, collecting the times of accessing the pushing main body by the sample user within a third preset time before the current pushing as the user characteristics of the sample user;

the pushing system can acquire the number of times that the sample user accesses the pushing subject within a third preset time before the current pushing as the user characteristic of the sample user. The third preset time length may be a time length preset according to needs, for example, set to be one month, and then the push system collects the number of times that the sample user accesses the push subject in one month before the current push. It should be noted that, the pushing system may record the operation of each user accessing the pushing subject, and count the number of access operation records to obtain the number of times that the sample user accesses the pushing subject; the third preset time period may be the same as or different from the second preset time period.

Step S602, collecting a time interval between an access time and a current push time as a user characteristic of the sample user, where the access time is an access time for the sample user to access the push subject within the third preset time before the current push, or an access time for the sample user to access the push subject last time before the current push.

The pushing system can acquire the time interval between the access time of the sample user to the pushing subject and the pushing time of the current pushing as the user characteristics of the sample user. The access time can be the access time of the sample user accessing the pushing main body within a third preset time before the current pushing, if a plurality of access records exist, a plurality of access times are collected for a long time, the time interval between each access time and the pushing time is respectively calculated, and the plurality of time intervals are all used as the user characteristics of the sample user; alternatively, the access time may be the access time of the sample user that last accessed the push principal before the current push.

In an embodiment, the push agent may use only the number of visits in step S601 as the user characteristic of the sample user, may use only the time interval in step S602 as the user characteristic of the sample user, and may use both the number of visits in step S601 and the time interval in step S602 as the user characteristic of the sample user.

In addition, an embodiment of the present invention further provides a push time prediction apparatus, and with reference to fig. 5, the apparatus includes:

the first prediction module 10 is configured to predict, by using a frequency model, based on a target user characteristic of a target user and preset alternative push frequencies, to obtain a first prediction result corresponding to each of the alternative push frequencies, where the frequency model is obtained by training, in advance, using the user characteristic and the push frequency as input characteristics and using a state of interest of a user to a push subject as a tag;

a first determining module 20, configured to determine a target push frequency from the alternative push frequencies according to the first prediction result, and determine a push period according to the target push frequency;

a second prediction module 30, configured to predict, by using a time model, based on the target user characteristic, preset push content, and each candidate time in the push cycle, to obtain a second prediction result corresponding to each candidate time, where the time model is obtained by training, with a user characteristic, push content, and push time as input characteristics in advance, and with a state of interest of the user to the push subject as a tag;

a second determining module 40, configured to determine, according to the second prediction result, a target push time for pushing the preset push content to the target user within the push cycle from each of the candidate times.

Further, the alternative push frequency is characterized as a vector, elements in the vector respectively correspond to each time period within a preset time period and are used for characterizing whether push is required within the corresponding time period, the first prediction result represents a probability of canceling attention of the target user to the push subject when the target user is pushed according to the corresponding alternative push frequency, and the first determining module 20 is further configured to:

Further, the first determining module 20 is further configured to:

Further, the preset push content includes a plurality of alternative push contents, and the second prediction module 30 is further configured to:

the second determining module 40 is further configured to:

Further, the apparatus further comprises:

a third determining module, configured to determine a sample user from currently pushed users after each pushing, where the sample user is at least from a pushed user who cancels attention to the pushing subject within a first preset duration after the pushing, and a pushed user who does not cancel attention to the pushing subject and has an access record to the pushing subject within the first preset duration after the pushing;

the first generation module is used for acquiring user characteristics of the sample user and pushing frequency of the sample user in a second preset time before the current pushing as input characteristics, and generating first sample data corresponding to the sample user by taking a concerned state of the sample user to the pushing main body in the first preset time after the current pushing as a label;

and the first training module is used for training the first sample data corresponding to each sample user to obtain the frequency model.

Further, the apparatus further comprises:

the second generation module is used for acquiring the user characteristics of the sample user, the push time of the current push and the push content of the current push as input characteristics, taking the attention state of the sample user to the push subject within the first preset time after the current push as a label, and generating second sample data corresponding to the sample user;

and the second training module is used for training by adopting the second sample data corresponding to each sample user to obtain the time model.

Further, the first generating module is further configured to:

The specific implementation of the push time prediction apparatus of the present invention has basically the same extension as that of each embodiment of the push time prediction method, and is not described herein again.

Furthermore, an embodiment of the present invention further provides a computer-readable storage medium, where the storage medium stores a push time prediction program, and the push time prediction program, when executed by a processor, implements the steps of the push time prediction method as described below.

The invention also proposes a computer program product comprising a computer program which, when executed by a processor, implements the steps of the push time prediction method as described above.

The embodiments of the push time prediction apparatus, the computer-readable storage medium, and the computer program product of the present invention may refer to the embodiments of the push time prediction method of the present invention, and are not described herein again.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.

The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims

1. A push time prediction method, the method comprising the steps of:

2. The push time prediction method of claim 1, wherein the alternative push frequency is characterized as a vector, elements in the vector respectively correspond to time periods within a preset time period and are used for characterizing whether push is required within the corresponding time period, the first prediction result represents a probability of a target user paying attention to the push subject when pushing to the target user according to the alternative push frequency,

3. The push time prediction method of claim 2, wherein the step of determining a push period according to the target push frequency comprises:

4. The push time prediction method according to claim 1, wherein the preset push content includes a plurality of candidate push contents, and the step of predicting, by using a time model, based on the target user characteristic, the preset push content, and each candidate time in the push cycle to obtain a second prediction result corresponding to each candidate time includes:

5. The push time prediction method according to any one of claims 1 to 4, wherein before the step of predicting by using the frequency model based on the target user characteristic of the target user and the preset alternative push frequencies to obtain the first prediction result corresponding to the alternative push frequencies, the method further comprises:

6. The push time prediction method of claim 5, wherein after the step of determining the sample user from the current pushed users after each push of the content, respectively, further comprises:

7. The push time prediction method of claim 5, wherein the step of collecting the user characteristics of the sample user comprises:

8. A push time prediction apparatus, comprising:

9. A push time prediction device, characterized in that the push time prediction device comprises: memory, processor and a push time prediction program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the push time prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a push time prediction program which, when executed by a processor, implements the steps of the push time prediction method according to any one of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, realizes the steps of the push time prediction method according to any of the claims 1 to 7.